Re: [PATCH v8 5/6] x86/signal: Detect and prevent an alternate signal stack overflow

[Borislav Petkov]

On Wed, May 12, 2021 at 10:55:04PM +0200, Thomas Gleixner wrote:
> On Wed, May 12 2021 at 18:48, Chang Seok Bae wrote:
> > On May 11, 2021, at 11:36, Borislav Petkov <bp@xxxxxxxxx> wrote:
> >> 
> >> I clumsily tried to register a SIGSEGV handler with
> >> 
> >>        act.sa_sigaction = my_sigsegv;
> >>        sigaction(SIGSEGV, &act, NULL);
> >> 
> >> but that doesn't fire - task gets killed. Maybe I'm doing it wrong.
> >
> > Since the altstack is already overflowed, perhaps set the flag like this -- not
> > using it to get the handler:
> >
> > 	act.sa_sigaction = my_sigsegv;
> > +	act.sa_flags = SA_SIGINFO;
> > 	sigaction(SIGSEGV, &act, NULL);
> >
> > FWIW, I think this is just a workaround for this case; in practice, altstack is
> > rather a backup for normal stack corruption.
> 
> That's the intended usage, but it's not limited to that and there exists
> creative (ab)use of sigaltstack beyond catching the overflow of the
> regular stack.

Right, with the above sa_flags setting (SA_ONSTACK removed) it does run the
SIGSEGV handler:

# [NOTE]        the stack size is 2048, AT_MINSIGSTKSZ: 3632
TAP version 13
1..3
ok 1 Initial sigaltstack state was SS_DISABLE
# sstack: 0x7f4e2e4d1000, ss_size: 2048
# [NOTE]        sigaltstack success
# [NOTE]        Will mmap user stack
# [NOTE]        Will getcontext
# [NOTE]        Will makecontext
# [NOTE]        Will raise SIGUSR1
# [NOTE]        signal SEGV
^^^^^^^^^^^

# [NOTE]        Will sigaltstack
ok 2 sigaltstack is still SS_AUTODISARM after signal
# Planned tests != run tests (3 != 2)
# Totals: pass:2 fail:0 xfail:0 xpass:0 skip:0 error:0

and exits normally. dmesg has:

[220514.661048] signal: get_sigframe: nested_altstack: 0, sp: 0x7ffc2846bca0, ka->sa.sa_flags: 0xc000004
[220514.661058] signal: get_sigframe: SA_ONSTACK, sas_ss_flags(sp): 0x0
[220514.661061] signal: get_sigframe: sp: 0x7f4e2e4d1800, entering_altstack
[220514.661064] signal: get_sigframe: nested_altstack: 0, entering_altstack: 1, __on_sig_stack: 0
[220514.661067] signal: sas[77819] overflowed sigaltstack

so at least we've warned that we've overflowed the sigaltstack.

[220514.661072] signal: get_sigframe: nested_altstack: 0, sp: 0x7ffc2846bca0, ka->sa.sa_flags: 0x4000004
[220514.661075] signal: get_sigframe: nested_altstack: 0, entering_altstack: 0, __on_sig_stack: 0

So I'm not even going to think about claiming that this is taking care
of the other productive ways of (ab)using the sigaltstack contraption
but from where I'm standing, it is not making it worse, AFAICT.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette